Long-term Renal Prognosis of Diarrhea-Associated Hemolytic Uremic Syndrome:  A Systematic Review, Meta-analysis, and Meta-regression FREE

The hemolytic uremic syndrome (HUS) is a disorder characterized by acute hemolytic anemia, thrombocytopenia, and renal insufficiency. HUS, especially in children, is a main cause of acute renal failure worldwide. The incidence of HUS is increasing,¹ with current estimates of 1 case per 50 000 patient-years for those younger than 18 years.² Ninety percent of childhood cases of HUS are associated with diarrhea and gastroenteritis and are due to Shiga toxin–producing Escherichia coli.³ This typical form of diarrhea-associated HUS causes toxin mediated vascular endothelial cell damage and in the kidney causes thrombotic occlusion of capillary lumens, glomerular endothelial cell swelling, apoptosis of glomerular and tubular cells, and extensive cortical necrosis.⁴

With improved recognition and supportive care, more patients are surviving the acute phase of diarrhea-associated HUS. An accurate estimate of the long-term renal prognosis is critical for patient counseling, follow-up, and monitoring. It informs the feasibility of future clinical trials, and guides the necessity of screening after large population outbreaks of Shiga toxin–producing E coli gastroenteritis. However, the long-term renal prognosis of diarrhea-associated HUS remains controversial, with markedly different results reported in various studies. This review was conducted to better quantify prognosis and to identify reasons for different estimates provided in the world's literature.

The primary questions of this review were: (1) What is the incidence of death or end-stage renal disease (ESRD) or a glomerular filtration rate (GFR) lower than 80 mL/min per 1.73m², hypertension, or proteinuria at least 1 year after diarrhea-associated HUS; (2) Which factors are associated with a worse long-term prognosis in individual studies; and (3) Which factors (such as secular differences, methodological quality, baseline patient characteristics, and initial treatments) are associated with a worse long-term prognosis at the study level. The last question was tested in meta-regression and the a priori hypotheses were a worse long-term prognosis would be evident in more recent studies (more children surviving severe disease), studies with longer follow-up, studies with more loss to follow-up,^5- 8 and studies with more patients with severe acute illness. Secondary questions included (1) Are patients with a less severe initial infection still at appreciable risk of long-term renal disease, and (2) Do patients develop renal disease after apparent renal recovery from diarrhea-associated HUS.

Case series, cohort studies, and randomized controlled trials were included if 2 among the reviewing group (A.X.G, R.S.S., F.R., D.M., M.P.R., M.S., or W.F.C.) had agreed independently that an article (1) described a study population of 10 or more patients with primary diarrhea-associated HUS (idiopathic or infection associated), (2) followed up patients for at least 1 year (because the natural history of the disease is an improvement of GFR over the first year^9- 10), and (3) reported 1 or more renal outcomes (proteinuria, hypertension, renal insufficiency, or ESRD). A third reviewer (A.X.G., R.S.S., or D.M.) resolved disagreements about whether a study should be included.

An independent review of citations from MEDLINE (OVID 1966 to February 2003) and Experta Medica (EMBASE, 1980 to February 2003) bibliographic databases was conducted (A.X.G. and F.R.). Full-text articles were retrieved if any of the authors considered any citation potentially relevant. The search strategy, developed with an experienced librarian, used terms most sensitive for identifying studies of prognosis¹¹ and was pilot tested and modified with known relevant articles. The search strategy included the terms hemolytic-uremic syndrome, purpura, thrombotic thrombocytopenic, Escherichia coli O157, longitudinal studies, kidney diseases, hypertension, and proteinuria (complete strategies available on request). Supplementary methods of finding studies included a review of relevant article bibliographies, a review of articles citing relevant articles in the Institute for Scientific Information index, a review of American Society of Nephrology meeting abstracts, and information provided by primary study authors.

Two of the reviewing group, using created forms, independently abstracted data on study and patient characteristics, renal measures, outcomes, and prognostic features. Disagreements were resolved by a third reviewer or by consensus. Non-English/non-French articles were an exception to this process, and a single reviewer (A.X.G.), with the help of a language translator, abstracted necessary data from Dutch, German, Japanese, Polish, Portuguese, and Spanish articles. Attempts were made to contact primary authors of all relevant studies to confirm the accuracy of abstracted data and to provide additional missing data.

The primary outcomes of this review were death or permanent ESRD, and a GFR lower than 80 mL/min per 1.73 m², hypertension, or proteinuria at follow-up. For one study,¹² original data provided by the study author was categorized according to criteria for this review. Confidence intervals (CIs) for single proportions were derived using the Wilson score method.¹³ We used χ² tests to assess between-study heterogeneity. An approach based on generalized estimating equations, which accounted for the within-study and between-study variability (random effects modeling), was used to derive pooled estimates of proportions and their variances.¹⁴ Estimates were computed using Excel. Overall, each study contributed a weight between 1% and 9% for any estimate. For each study, we assessed 8 measures of quality. We selected a priori not to use an existing quality scale or to create our own scale. We assessed 2 of the most important measures in the meta-regression analyses—percentage lost to follow-up and whether the method of renal assessment was reported.

In meta-regression, the following prognostic factors were considered at the study level: the mean age of patients; duration of follow-up; the mid point year of the study; whether the method of renal assessment was reported; the proportion of patients who presented with central nervous system (CNS) symptoms (coma, seizures, or stroke); the proportion of patients who received acute dialysis, plasma infusion or exchange, or corticosteroids; and the proportion who were lost to follow-up. Additionally, for the outcomes of GFR lower than 80 mL/min per 1.73 m², hypertension, or proteinuria, the proportion of patients experiencing the competing event of death or permanent ESRD was considered. To examine the impact of these factors on study outcomes, exploratory meta-regressions were conducted using logistic normal random effects models using SAS PROC NLMIXED (SAS Inc, Cary, NC). Values for some factors were not available for some studies. For each meta-regression, only studies for which all factors were available were included in the analysis. The explanatory ability of each meta-regression model was summarized as the proportion of between-study variability explained on the logit scale. Univariate meta-regression was used for each outcome and each factor. Factors found to be significant at the P = .10 level were included in multivariate meta-regression models. Factors found to be nonsignificant in a multivariate model were subsequently removed. To generate best-fit lines in meta-regression graphs, an approximate correction was used to convert conditional means from the logistic normal model into marginal mean curves.¹⁵

From screening more than 3384 citations, 124 full-text articles were retrieved, and 49 articles were included in this review. The agreement beyond chance between 2 independent reviewers in the reviewing group for citation screening, article inclusion, and data abstraction was good (κ range, 0.56-0.74 on 20 different measures). The reasons for article exclusion were the study's follow-up was less than 1 year, reporting of renal outcomes was unclear, selection of population for follow-up was nonrepresentative, inception of a cohort of patients with HUS was unclear, population included a large proportion of recognized secondary causes of HUS, or patients seemed to have been described in other articles already included in this review. (Excluded articles are available on request.) A single-adult study was excluded from this review.¹⁶ Of the 49 articles, we contacted 45 primary authors, 25 of whom provided additional information and confirmed the accuracy of abstracted data. All included studies were considered representative of patients with typical diarrhea-associated HUS. Results of patients with recognized secondary forms of HUS were excluded when data were abstracted and analyzed. From solely reading the primary reports, we found occasional ambiguity about the reported cause of HUS. Exclusion of 9 studies with such ambiguity from our analyses did not change the summary estimates provided in this review.^10,17- 24

The 49 studies included 3476 patients living in 18 different countries who were followed up for a mean (SD) of 4.4 (4.2) years. The mean range of follow-up was 1 to 22 years. The studies were conducted between 1950 and 2001 (Table 1, Table 2, and Figure 1).^{5- 10,12,17- 58} Fifty-two percent of patients were female, and the mean age was 2.4 years (range, 0.1-18 years at recruitment into studies). Seven studies were associated with outbreaks of Shiga toxin-producing E coli, the source was uncertain for 2 studies,^28,38 and the others were associated with municipal water,⁵⁸ radish sprouts,⁵⁷ hamburgers,⁸ raw ground beef,^12,59 and fermented sausage.⁵⁶ During the last 15 years, identification of Shiga toxin–producing E coli in study reports did not appreciably improve. With the exception of outbreaks, only 14% of all patients had E coli O157:H7 confirmed by either stool culture, antibody-to-stool verotoxin, or serological testing.

Three study designs were clinical trials of urokinase and heparin⁴⁷ and plasma exchange or infusions^50- 51 with the remaining being case-series and cohort studies. (We considered nonrandomized trials with historical control groups to be cohort studies [Table 1].) Excluding the 3 clinical trials from this review did not change the main study results. When assessing quality, 43% were prospective, 73% described consecutive HUS patients at an institution, 56% defined cutoff values for the diagnosis of HUS, 71% described important baseline characteristics (such as demographic and exposure characteristics), 69% described treatments used during acute HUS, 16% described treatments used during follow-up, 61% used clear methods for defining and reporting renal outcomes, and 47% measured all patients at the same follow-up time or used statistical methods that adjusted for varying lengths of follow-up. On average, 21% of patients were lost to follow-up (range, 0% to 59%), and 64% of studies either had less than 10% lost to follow-up or described the characteristics of those lost to follow-up.

When described, proteinuria, hypertension, low GFR, and combinations of these were defined and measured in different ways in the primary studies. Different definitions of proteinuria used in the primary studies included a random urine dipstick (albustix, labstix, multistix) of at least 0.1 g/L (trace, 0.3 g/L [1⁺] or 1.0 g/L [≥2⁺]),^9,23 or an early-morning urine protein-to-creatinine ratio of at least 177 mg/g (≥20 mg/mmol) ^5,38 or at least 265 mg/g (≥30 mg/mmol),⁵⁶ or a random urine albumin-to-creatinine ratio of at least 177 mg/g (≥20 mg/mmol),⁵⁸ a 24-hour urine protein of at least 100 mg/d, 150 mg/d,^18,20,24 200 mg/d,^25,42 250 mg/d, or 300 mg/d,^23,31,44 or a 24-hour urine protein of at least 100 mg/m² of body surface area.^33,35 Definitions of hypertension included the use of antihypertensive medications,^27,54 1 blood pressure measurement higher than the 90th,⁶ 95th,^{9,12,25,33,42,50} 97th, or 99th⁸ percentile, or 10 mm Hg above the 95th²³ or 97th percentile,⁵¹ using various population norms defined by age,^17,39 sex,⁵⁶ weight and/or height.⁵ Definitions of decreased GFR included an elevated serum creatinine level, GFR estimated from predictive equations,⁶⁰ 24-hour urine creatinine clearance, measured GFR using injected inulin,²³ iothalamate,⁸ technetium diethylenetriamine pentaacetic acid,⁷ or EDTA,^9,41 with usual cutoff points of abnormality of less than 80 mL/min per 1.73 m² of body surface area.⁹ In this review, the expression long-term renal sequelae refers to a low GFR (usually <80 mL/min per 1.73m²), hypertension, or overt proteinuria at last follow-up, as assessed in the individual studies. Similarly, renal recovery refers to a normal GFR (usually ≥80 mL/min per 1.73m²) and no evidence of hypertension or overt proteinuria as assessed in the individual studies.

For all outcomes, the variability between studies was larger than would be expected by chance (significant heterogeneity, P<.001). Given the large between-study variability with wide ranging heterogeneity (Figure 1), mathematically pooled results should be interpreted with caution. However such results are provided because they represent current best evidence for clinical care and guide sample-size calculations for future epidemiological studies. Death or permanent ESRD ranged from 0% to 30% (death, 0%-23%; permanent ESRD, 0%-17%; Table 3 and Figure 1). The mathematical pooled estimate of the incidence of death or permanent ESRD was 12% (95% confidence interval [CI], 10%-15%; death 9%, [95% CI, 7%-11%; permanent ESRD, 3% [95% CI, 2%-5%]), with the majority of cases occurring during the acute phase of HUS.

At a minimum of 1 year of follow-up, 2372 patients without permanent ESRD had an assessment of renal function. Renal sequelae ranged from 0% to 64% (Table 3 and Figure 1). The majority of studies did not differentiate whether these were persistent findings after acute HUS or whether they developed after apparent initial renal recovery. The pooled estimate of renal sequelae was 25% (95% CI, 20%-30%). The pooled estimate of the incidence of a GFR of 60 to 80 mL/min per 1.73 m² was 8% (95% CI, 5%-11%); of 30 to 59 was 6% (95% CI, 3%-8%); of 5 to 29 was 1.8% (95% CI, 0.8%-3%), hypertension was 10% (95% CI, 8%-12%), and proteinuria was 15% (95% CI, 10%-20%).

A number of prognostic features were described in the individual studies. Many of the primary study authors described an association between an increased severity of acute illness (greater infection or host response) and worse long-term prognosis, including an elevated white blood cell count higher than 20 × 10³/µL with neutrophilia,^29,38 a high serum creatinine or urea concentration,⁵³ central nervous system symptoms (reduced consciousness, coma, stroke, or seizures),^{10,23,33,36,41,44,49,53} ischemic colitis,^33,49 and hypertension.^{9- 10,23,36,44,48,53} Lower and prolonged levels of anemia and thrombocytopenia were inconsistently associated with both worse⁸ and improved³⁸ outcomes. Compared with patients with oliguria of 8 days or less (an approximate urine output less than 300 mL/m² per day), those with oliguria of greater than 8 days or anuria of 1 to 8 days, and those with anuria of greater than 8 days had a step-wise worsening of prognosis.^{9,18,20,22- 23,26,33,46,48} A longer duration of dialysis was associated with worse prognosis, and no patient achieved full renal recovery when dialysis therapy exceeded 4 weeks.^{5,23,29,41- 42} Similarly after 3 weeks, the longer the duration of hospitalization, the worse the outcome.⁵⁶

Certain studies examined the significance of renal biopsy near acute illness. Those with cortical necrosis or arterial thrombotic microangiopathy demonstrated a worse prognosis compared with patients with isolated glomerular lesions.^23,25,46 In patients with isolated glomerular lesions, those with more than 50% glomerular involvement had a worse prognosis.⁴⁶

Younger ages (on average <2-5 years old)^10,24,31,33 and female sex³⁸ had an inconsistent effect on results. One study suggested a worse prognosis with HUS during winter.³⁸ Increased C-reactive protein⁵³ and cytokine levels (IL-6, IL-10, IL-1, Ra)²⁴ were associated with worse prognosis.

With respect to treatment, living no further than 100 km (62.5 miles) from a tertiary center,⁴² and early recognition and treatment including dialysis^10,42 were associated with better prognosis. Better outcomes were not consistently achieved with streptokinase^18,43 or plasma exchange.^{21,40,50- 51}

At the study level, severity of acute illness was associated with worse long-term prognosis in both univariate and multivariate analyses. Studies with a higher proportion of patients with CNS symptoms (coma, seizures, or stroke) had a higher proportion of patients who died or who developed permanent ESRD, explaining 44% of the between-study variability (P = .01; Figure 2). Similarly, studies with a higher proportion of patients requiring acute dialysis had a higher proportion of patients who died or developed permanent ESRD, explaining 10% of the between-study variability (P = .02), and had a higher proportion of patients with long-term renal sequelae, explaining 15% of the between study variability (P<.001; Figure 3).

Study level factors such as a higher proportion of patients of older age, female sex, receiving corticosteroids, and factors such as the years in which the study was conducted and whether the study reported careful methods of renal function assessment were not associated with worse outcomes. Similarly, studies with a lower proportion of patients with death or ESRD did not demonstrate a higher proportion of patients with renal sequelae at follow-up. In univariate but not multivariate analyses, studies with a longer follow-up time demonstrated a higher proportion of patients with long-term renal sequelae.

However, in both univariate and multivariate analyses, studies with a higher proportion of patients lost to follow-up demonstrated a higher proportion of patients with long-term renal sequelae, explaining 28% of the between-study variability (P = .001; Figure 4). When the analysis was limited to the 30 studies with better than 90% follow-up, with a mean follow-up of 3.6 years (range, 1-10 years), the pooled estimate of death or permanent ESRD was 11% (95% CI, 8%-14%) and renal sequelae was 17% (95% CI, 12%-22%). Studies with a higher proportion of patients receiving plasma infusion or exchange had a lower proportion of patients who died or developed permanent ESRD, explaining 8% of the between study variability (P = .03), and a lower proportion of patients with renal sequelae, explaining 28% of the between-study variability, P = .03; Figure 5). Visually this trend seemed to be influenced by a small number of studies.^45,52,54

A few studies have suggested that patients with less severe forms of HUS, including those with a preserved urine output, may still demonstrate renal sequelae at follow-up.^23,25 For example, 2 of 5 patients with acute diarrhea-associated HUS and mild renal impairment (normal urine output, no dialysis) demonstrated hypertension and proteinuria at long-term follow-up.²⁵ Similarly 5 of 18 patients with acute diarrhea-associated HUS and normal urine output later developed long-term renal sequelae including chronic renal failure.²³

In studies of patients who develop long-term renal sequelae, the majority of studies did not differentiate patients who apparently completely recovered after the acute illness from those who demonstrated persistent renal abnormalities. Four studies that did differentiate these states using a single-screening test of clearance suggested that 8% to 61% of those who seemed to have a normal GFR after diarrhea-associated HUS (assessed either by a normal serum creatinine concentration⁷ or a calculated³³ or measured GFR >80 mL/min per 1.73 m²)^5,9 went on to develop a GFR lower than 80 mL/min per 1.73m², hypertension, or proteinuria during long-term follow-up. Similarly, a single study suggested that a quarter of those who recovered with an absence of proteinuria (<250 mg of protein per day) went on to develop renal sequelae during long-term follow-up.³³ On the other hand, 4 studies that considered 3 measures of renal health in combination suggested that no one who recovered with an absence of hypertension, an absence of dipstick proteinuria, and a calculated or measured GFR higher than 80 mL/min per 1.73m² developed future renal sequelae,^5,22- 23,26 recognizing these studies are limited by follow-up times of less than 5 years.

Studies of diarrhea-associated HUS are highly variable in their methodological rigor, methods of renal assessment, and estimates of long-term prognosis. For this reason, the long-term renal prognosis of diarrhea-associated HUS remains controversial. As highlighted in this review, the incidence of death or permanent ESRD ranged from 0% to 30% in the primary studies, and 0% to 65% of patients demonstrated long-term renal sequelae. Our mathematical pooled estimates based on existing studies suggest an average of 4 years after experiencing diarrhea-associated HUS, 9% of patients die (most during the acute phase of illness), an additional 3% develop permanent ESRD, and 25% demonstrate renal sequelae. Severity of the acute illness, especially the presence of CNS symptoms, is strongly associated with worse long-term outcome. These symptoms of coma, seizures, or stroke may be due to ischemic thrombotic microangiopathy, concurrent hypertension, or metabolic alterations of hyponatremia and hypocalcemia.⁶¹ The need for initial dialysis was also strongly associated with worse outcome. No one achieved full renal recovery when dialysis therapy exceeded 4 weeks.

Some studies have confirmed long-term sequelae even after milder forms of diarrhea-associated HUS.^22,24 The natural history of diarrhea-associated HUS is often an improvement of GFR (approximately 25 mL/min per 1.73m² over the first year^9- 10). Thus it seems prudent to screen patients at least once for silent renal disease a year after acute diarrhea-associated HUS. As the sole method of characterizing patients free of disease, a nuclear GFR higher than 80 mL/min per 1.73m² after recovery is an inadequate screening test. However, those who demonstrate a predicted creatinine clearance higher than 80 mL/min per 1.73m², ⁶⁰ no overt proteinuria (≥1⁺ or 0.3 g/L on dipstick), and no hypertension (based on population reference values), a potentially easily implemented cost-effective screening strategy, appear to have an excellent prognosis during early follow-up.^5,22- 23,26

Similar to other renal insults, it is theorized that diarrhea-associated HUS can lead to a critical reduction in nephron number, with unsustainable remnant single-nephron hyperfiltration⁶² and progressive renal disease.⁶³ However, these results confirm that the relationship between longer follow-up time and worse prognosis is in part complicated by a high number of patients lost to follow-up in some studies. Patients lost to follow-up contribute to worse estimates of long-term prognosis because they are typically healthier than those followed up. Further prognostic studies should adhere to the methodological quality criteria outlined in this review, and particularly should strive for less than 10% of patients lost to follow-up. Different definitions for surrogate outcomes of proteinuria, low GFR, and hypertension performed at different follow-up times complicate the interpretation and utility of results from many of the primary studies and the summary results of this review. A uniformly adopted time and method of measurement, with specified cutoff levels that have been associated with death or ESRD in prospective studies of children,⁶⁴ would be advantageous. Further characterization of prognostic factors and novel methods of screening patients after diarrhea-associated HUS, although critical for clarifying pathophysiology,^6,9,65- 66 may prove difficult to apply clinically. Rather, improving the detection of Shiga toxin–producing E coli as the cause of diarrhea-associated HUS, which has been limited even in studies during the last 15 years, seems necessary for the proper characterization of future cohorts. In addition, given that long-term sequelae have been described after milder forms of diarrhea-associated HUS,^23,25 research examining the incidence of renal disease after toxigenic E coli gastroenteritis seems essential.

Future therapeutic studies may consider a preventive approach to the medical consequences of Shiga toxin–producing E coli, including increased sanitation, surveillance, assessment, reporting, and possibly bovine and human vaccination.^67- 68 Avoidance of antibiotics and antimotility agents during gastroenteritis^69- 70 and the role of protective antibodies⁷¹ and oral synthetic verotoxin-receptor analogs attached to chromosorb ⁷² remain to be clarified. Early recognition and referral to a specialized medical care center experienced in the critical care of acute diarrhea-associated HUS seems to be a particularly important strategy.⁴² Corticosteroids, antiplatelet agents, anticoagulant therapy, and thrombolytics during the acute phase of diarrhea-associated HUS are not clearly effective.^47,73 Careful blood pressure control and renin-angiotensin system blockade may be particularly beneficial for those who demonstrate renal sequelae after diarrhea-associated HUS. With an event rate of 12% death or permanent ESRD at 4 years, this review confirms that multicenter, multinational, clinical trials with thousands of patients are required for adequate statistical power to demonstrate the benefits of any therapy on these outcomes.

A therapy that may warrant this level of attention in future clinical trials is plasma exchange. Sixteen studies included in this review described a portion of patients treated with infusion or exchange. However smaller and possibly underpowered trials of plasma have not shown efficacy in diarrhea-associated HUS.^49- 50 These meta-regression results of the potential benefits of plasma are intriguing, recognizing their interpretation is complicated by their observational nature, analytic limitations,⁷⁴ trends influenced by a small number of studies,^45,52,54 and an absence of risk benefit assessment. Publication biases (in which studies with larger efficacy are more likely to be reported) may also exist.

In summary, death or ESRD occurs in about 12% of patients 4 years after diarrhea-associated HUS, and 25% of survivors demonstrate long-term renal sequelae. The severity of acute illness, particularly CNS symptoms and the need for initial dialysis, is strongly associated with a worse long-term prognosis. Patients lost to follow-up contribute to worse estimates in some studies.